Compressive full-Stokes spectropolarimetric imaging (SPI), integrating passive polarization modulator (PM) into general imaging spectrometer, is powerful enough to capture high-dimensional information via incomplete measurement; a reconstruction algorithm is needed to recover 3D data cube (x, y, and λ) for each Stokes parameter. However, existing PMs usually consist of complex elements and enslave to accurate polarization calibration, current algorithms suffer from poor imaging quality and are subject to noise perturbation. In this work, we present a single multiple-order retarder followed a polarizer to implement passive spectropolarimetric modulation. After building a unified forward imaging model for SPI, we propose a deep image prior plus sparsity prior algorithm for high-quality reconstruction. The method based on untrained network does not need training data or accurate polarization calibration and can simultaneously reconstruct the 3D data cube and achieve self-calibration. Furthermore, we integrate the simplest PM into our miniature snapshot imaging spectrometer to form a single-shot SPI prototype. Both simulations and experiments verify the feasibility and outperformance of our SPI scheme. It provides a paradigm that allows general spectral imaging systems to become passive full-Stokes SPI systems by integrating the simplest PM without changing their intrinsic mechanism.
Classification is the focus and difficulty of hyperspectral imaging technology. Hyperspectral data have twodimensional spatial information and one-dimensional spectral information, which are presented as three-dimensional data blocks with large amount of information, meanwhile high-dimension, high nonlinearity and limited training samples bring great challenges. Deep learning can extract and analyze the features of target data step by step by building multi-layer deep nonlinear structure. The advanced feature, multi scale abstract information extracted by convolution neural network applied to image processing can improve the classification accuracy of complex hyperspectral data. We regard pixel level hyperspectral classification as semantic segmentation network, and creatively introduce squeeze-and-excitation network and pyramid pooling network into hyperspectral classification network and proposed a model based on the structure of 2D-3D hybrid convolution neural network, it can learn deeper spatial spectral features and fusion to improve the accuracy and speed of hyperspectral classification.
Usually, the practical analysis states of an imaging polarimeter needs to be calibrated, with a set of standard polarization states, for the accurate reconstruction of Stokes parameters. However, it is really challenged to get the standard elements over wide field of view (FOV), broad waveband, large aperture, or other non-trivial conditions. Even if the system is well calibrated, the calibrated system will be disturbed in the vibration environment. To avoid the difficult from the standard polarization states, an iterative reconstruction method is presented at the first time to recover the polarization parameters from the data acquired by linear-Stokes polarimeters without polarimetric calibrations. Inspired from phase shifting interferometry, the method employs two least-squares iterative procedure and requires no any extra element for assistant. And we extend the method to a channeled linear imaging spectropolarimeter, channeled linear imaging spectropolarimeter can measure a two-dimensional distribution of spectrally-resolved linear Stokes parameters in a single-shot polarization modulation. However, the state-of-art reconstruction method, Fourier transform method (FTM), usually transforms the modulated spectrum into the frequency domain for further processing. As a result, there is channel crosstalk issue that limits available frequency bandwidth. In addition, FTM needs extra phase calibration to decode final spectra. We present a continuous slide iterative method (CSIM) in the spectral domain to avoid the use of the Fourier transform and phase calibration. It combines a sliding unit cell kernel in the spectral domain that provides unit cell tracking and a loop of twostep least-squares fit that estimates spatially-resolved polarized spectra.
The use of wind imaging interferometer to retrieve wind field information requires high accuracy of the instrument in phase shift. Traditional wind field inversion algorithms invert wind field information such as wind speed and temperature based on equal phase-shift interferometers. In actual situations, the phase shift of the wind imaging interferometer does not perfectly meet the design requirements, producing inversion result errors. In this paper, the inversion algorithm of an arbitrary phase shift wind imaging interferometer is studied, and the feasibility of the AIA (advanced iterative algorithm) inversion algorithm is verified under the condition of low accuracy of the instrument phase shift. The comparison of the inversion results of the AIA algorithm in wide-field interference and non-widefield interference are respectively discussed. The influence of the unevenness of the strip system caused by the temperature distribution on the inversion accuracy of the AIA algorithm is analyzed. The conclusions of this paper can provide theoretical support for phase-shifting calibration of wind imaging interferometers and wind field information inversion in low-precision instrument.
Change detection (CD) is the process of identifying differences in the state of an object or phenomenon by observing it at different times. CD is one of the earliest and most important applications of remote sensing technology. The hyperspectral image (HSI) of remote sensing satellite provides an important and unique data source for CD, but its high dimension, noise and limited data set make the task of CD very challenging. Traditional algorithms are no longer suitable for hyperspectral data processing. Recently, the success of deep convolutional neural networks (CNN) has widely spread across the whole field of computer vision for their powerful representation abilities. Therefore, this paper combines traditional algorithms and deep learning techniques to solve the CD task of hyperspectral remote sensing images. The proposed two-branch Unet network with feature fusion (Unet-ff) model in this paper uses neural networks to automatically extract features to achieve end-to-end change information detection. In order to improve the degree of automation in the application, we select the most effective results as the training sample for the neural network which obtained by various traditional algorithms, and use ground truth to evaluate the detection results. For the characteristics of hyperspectral data, we use effective dimensionality reduction methods and rich data amplification methods to improve the detection accuracy. Experimental results show that our method can achieve better results on the existing classical datasets.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.